feat(patreon): seen-ledger table + model — ingester build step 2a (plan #697)

patreon_seen_media(source_id, filehash, post_id, seen_at), UNIQUE(source_id,
filehash) — our own queryable replacement for gallery-dl's archive.sqlite3.
Routine walks skip seen media; recovery mode bypasses the ledger. filehash is
a 32-hex CDN MD5 or a video:<post>:<media> sentinel (String(128)). alembic
0037 (← 0036). Integration test covers dedup + savepoint recovery.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_patreon_seen_media.py

@@ -0,0 +1,73 @@
+import pytest
+from sqlalchemy import func, select
+from sqlalchemy.exc import IntegrityError
+
+from backend.app.models import Artist, PatreonSeenMedia, Source
+
+pytestmark = pytest.mark.integration
+
+
+async def _source(db):
+    artist = Artist(name="Seen Ledger Artist", slug="seen-ledger-artist")
+    db.add(artist)
+    await db.flush()
+    source = Source(
+        artist_id=artist.id,
+        platform="patreon",
+        url="https://www.patreon.com/seenledger",
+    )
+    db.add(source)
+    await db.flush()
+    return source.id
+
+
+@pytest.mark.asyncio
+async def test_patreon_seen_media_dedup(db):
+    source_id = await _source(db)
+    db.add(
+        PatreonSeenMedia(
+            source_id=source_id,
+            filehash="0123456789abcdef0123456789abcdef",
+            post_id="123456",
+        )
+    )
+    await db.flush()
+
+    seen_at = await db.scalar(
+        select(PatreonSeenMedia.seen_at).where(
+            PatreonSeenMedia.source_id == source_id
+        )
+    )
+    assert seen_at is not None
+
+    # A second sighting of the same (source_id, filehash) is rejected by the
+    # unique constraint. Wrap in a savepoint so the IntegrityError doesn't
+    # poison the outer transaction.
+    with pytest.raises(IntegrityError):
+        async with db.begin_nested():
+            db.add(
+                PatreonSeenMedia(
+                    source_id=source_id,
+                    filehash="0123456789abcdef0123456789abcdef",
+                    post_id="654321",
+                )
+            )
+            await db.flush()
+
+    # Same filehash but a different post_id sentinel still inserts: the
+    # dedup axis is (source_id, filehash), not post_id.
+    db.add(
+        PatreonSeenMedia(
+            source_id=source_id,
+            filehash="video:123456:7890",
+            post_id="123456",
+        )
+    )
+    await db.flush()
+
+    count = await db.scalar(
+        select(func.count(PatreonSeenMedia.id)).where(
+            PatreonSeenMedia.source_id == source_id
+        )
+    )
+    assert count == 2